Affiliation
American Association of Variable Star Observers (AAVSO)
Thu, 05/01/2014 - 17:24
I have been making PEP observations since last July, and over the months I have attempted epsilonV calibrations on several of the red/blue star pairs on the AAVSO PEP calibration web page. I find myself getting some large differences (eg: 40%) from pair to pair, and I wanted to ask which star pairs people are successfully using, and how much variation in epsilonV you are finding from one calibration to the next.
Tom
A possible source of trouble is that most of the 12 red/blue pairs in the AAVSO eV calibration list contain, according to Simbad or VSX, at least one variable or suspected variable. Below is a summary:
Andromeda: blue star (HR 477)
Perseus: red star (HR 1052)
Serpens: blue star (HR 5859)
Hercules: red and blue stars (HR 6418, 6436)
Ophiuchus: blue star (HD 161261)
Aquarius: red star (HR 8453)
Pegasus: blue star (HR 8880)
Columba: blue star (HR 2265)
Hi Tom,
These are all low amplitude variables or suspects at best. I would assume any variations (where they exist) would not impact your measurements at that level, but I could be mistaken. Here's what I found from looking at the Simbad and VSX entries and from looking at the Hipparcos light curves (I have not checked for BSM photometry):
Andromeda: blue star (HR 477)
NSV star
All photometry within uncertainties in Hipparcos time series
Perseus: red star (HR 1052)
NSV star
All photometry within uncertainties in Hipparcos time series
Serpens: blue star (HR 5859)
Not an NSV star
Slight evidence of variability at the 0.02mag level in Hipparcos
Hercules: red and blue stars (HR 6418, 6436)
NSV star, variability uncertain
6418 slowly variable at the level of 0.01mag in Hipparcos
NSV star, variability uncertain
6436 slowly variable at the level of 0.01mag
Ophiuchus: blue star (HD 161261)
V2315 Oph, definite variable (ellipsoidal?)
Slight evidence of variability at the level of 0.03mag in Hipparcos
Aquarius: red star (HR 8453)
NSV star, variability uncertain
Slight evidence of variability at the level of 0.02mag in Hipparcos
Pegasus: blue star (HR 8880)
tau Peg, definite variable (dSct, 0.02mag)
Slight evidence of variability at the level of 0.02mag in Hipparcos
Columba: blue star (HR 2265)
NSV star (possible low amplitude dSct?)
Slight evidence of variability at the <0.02mag level in Hipparcos
Admittedly, I am new to this, but it seems like such small variations can be significant. Consider the Aquarius pair. On the PEP eV web page, the magnitude of the red star is listed as 6.015. But on the PEP eV archival chart, the magnitude is 6.005. The 5 mmag difference is enough to change my computed eV by 10%. << OOPS! TYPO: 1st magnitude should be 6.01 not 6.015 >>
It was my understanding that I should expect an eV for my instrument in the neighborhood of -0.050; is this correct? The archival magnitudes of the Aquarius pair gave -0.0448, while the web page magnitudes gave -0.0397. A later red/blue observation of the Andromeda pair gave -0.0449.
I'm not sure you how you're doing the calculation -- would you please email me your observations and calculations offline? I'd like to calculate your epsilon(V) using our copy of RBTRAN, which is what we do for most PEP observers.
In the epsilon(V) calculation using the Aquarius pair, a difference in catalog V values of 5 millimag should result in a difference in epsilon(V) of .005 if delta(B-V) is assumed to be -0.985. That will yield photometry with systematic offsets of less than 1 percent for nearly all stars in the PEP program.
And remember, your photometry of a red-blue pair will include uncertainty at the +/-0.005 level (if you observer on a very good photometric night). Many measures of the pair will beat down the uncertainty. Observations on additional nights will also improve the average.
But best of all, remember that the comp stars in the AAVSO PEP program are selected (in almost all cases) to match the variable as nearly as possible in color. A B-V color difference of 0.2 and an error in your epsilon-V of 0.01 will only add a transformation uncertainty of 0.002 magnitude to the variable observation.
I almost always used the red-blue pair in Leo, but it's not mentioned in your list.
The Leo Minor, Orion, Centaurus, and Sculptor pairs do not appear to have any variables. What eV values do you get with LMi?
Regarding color difference between variables and comps, the 5 recommended stars for new PEP observers have delta B-V values of 2.173 (CE Tau), 1.265 (RS Cnc), 0.38 (IN Hya), 0.608 (W Boo), and -0.517 (Rho Cas), most of which are significantly greater than 0.2 in absolute value.
The eps-V for my SSP-3 with the LMi pair was -0.36 (I believe it was negative; it's been more than a decade since I used my SSP-3, I've been using the SSP-5 in recent years). I tested it about every year and always got the same result, give or take a smidgen. I also tested it on different telescopes and got a very similar value - it would seem that the transformation coefficient is more dependent on the filter and detector than on the optical system.
Once or twice I tried other star pairs and got somewhat different results. I also did a all-sky determination and again got similar but somewhat different results. But all these determinations were the same within 0.01 magnitude. So I just decided to go with it. A B-V color difference between a variable and comp star and an error of 0.01 in eps-V is still going to produce a delta V within 0.02 of the ideal value. The AAVSO PEP program would welcome several thousand observations with that precision.
I can't follow your next-to-last sentence. How big a value for delta B-V?
Jim Fox's introduction to the AAVSO PEP program states: "Integrity of our database requires that observations be accurate to within about 0.02 magnitudes." It seems that the scenario you describe would put the observation at the limit of acceptable deviation - or am I misinterpreting?
Also, an eV of -0.36 seems huge; do you mean -0.036?
I purchased my SSP3 in 2014. Its epsilon V is -0.054 using the Hercules pair (using my C14). As a beginner photometrist I selected W Boo as one of my program stars. If you want to coordinate some times of observation we could determine if there are any systematic errors between our setups using this star. May be a bit of an issue with the difference in airmass for our observations as we are about 3 hours apart in longitude. I am also open to picking serveral non-variable stars to test for systematic errors between our setups. Let me know if you are interested in coordinating any observation. Maybe we can get some of the other single channel PEP observers intersested. A periodic comparison of our setups through observations would be useful.
The reason I chose the SSP3 was its accuracy, it would be great to know how my setup compares/transforms to others. I believe there are good bright program stars that could benefit from the precision of SSP3 setups. Unfortunately, with few single channel PEP observers the density of the observations can be a bit sparse.
Jim
Jim:
I am very interested in simultaneous observing sessions and would be glad to coordinate them. Feel free to contact me offline and we could go into details. I am planning a project this summer where three photometers would be used simultaneously from my location to see how well they can agree. As part of that project, I am trying to sort out the eV calibration variations I am experiencing.
I wanted to bring this discussion up-to-date with some new analysis and new data. Last year, I made eV measurements on the following PEP red/blue calibration pairs: Oph, Aqr, Peg, And, Per, Ori, and LMi. The calibration computation compares the standard delta V (Vblue-Vred) with the measured delta v (vblue-vred) of the blue and red stars. A small difference between the two is desired. I tried playing around with these values in the form of a percent error, 100*(delta_V - delta_v)/delta_V, and got some interesting results.
Percent Error
Star Pair
2013
2014
Perseus
3
Orion
6
Aquarius
15
2
Leo Minor
18
Pegasus
29
27
Ophiuchus
182
Andromeda
520
228
If one plots the logarithm of the 2013 percent error versus delta(V-R), one gets a fairly good linear fit for all the pairs except Pegasus (see attachment). For percent error versus delta(B-V), there is a good fit except for Pegasus and Ophiuchus. My interpretation is that as the color index difference goes linearly from more negative to less negative, the error increases exponentially. I am presently trying to figure out what may be going on with Pegasus and Andromeda that makes them outliers.
This year, I have begun making eV calibration measurements, again. So far, I have data for only the Aqr, Peg, and And pairs. These data were taken with a different photometer+filter combination (same telescope), but they show the same pattern of error.
Tom,
I made a measurement of the Pegasus pair last night. I used an IR blocking filter in addition to m V filter. Conditions were variable with some high cirrus moving in and out so I did get some scatter in the data. Because of the noise I am providing two values below, one is with the full 11 data points, and the other with 7 data points (4 'outliers' with high variance removed).
Data Set Epsilon V StdDev Tom's error metric
Full (11 Points) -0.057 0.019 30%
Outliers removed (7 Points) -0.048 0.011 25%
Over the past year I have used the Hercules pair for calibration and my epsilon v with that pair is -0.054. I have not done the hypothesis test, but my feeling is that the results above are not sufficient to reject the hypothesis that the epsilon V calculated with the Pegasus pair is the same as that calculated with the Hercules pair. (Although, I did not use an IR blocking filter on the Hercules pair). I will need a much better night to come to a definate conclusion.
The delta V for the And pair is only 0.01 so calculating epsilon V with this pair would be more susceptible to measurement noise.
If conditions are better tonight I will make an attempt at both Pegasus and Andromeda. If conditions are really good I will try the runs with and without the IR filter.
Jim
The Pegasus blue star is a variable, with a period of about 80 min (according to VSX).
I made a mistake calculating the Andromeda pair errors (punching too much data on a hand calculator). A revised table is below. The changes actually improve the linear fits. I have added delta V information.
Revised Error Percentages
Star Pair
2013
2014
| delta V |
Perseus
3%
1.46
Orion
6%
0.29
Aquarius
15%
2%
0.26
Leo Minor
18%
0.378
Pegasus
29%
27%
0.2
Ophiuchus
182%
0.51
Andromeda
320%
128%
0.01
I attach a revised error vs. delta(B-V) plot, and a plot of error vs. delta V.
I used the wrong delta(B-V) in the calculation of the epsilon V using the Pegasus pair. I used -1.055 instead of -0.44. (read one entry too far down in the table giving delta (B-V)). My revised numbers for epsion V for the 11 point and 7 point samples are -0.138 and -0.116. I performed a significance test on these values. As expected they are now statistically different from the value calculated using the Hercules pair (which was -0.054).
Jim
Maybe Pegasus is horsing around. Does anybody know of any recent photometry on SAO 91253 and SAO 91186? Maybe BSM? The catalog data for these stars agrees reasonable well with the AAVSO webpage on calculating epsilon V, but something seems off. The epsilon V calculated with these values is off by more than a factor of 2 from that calculated with other red blue pairs, when using an SSP3 photometer. Photometry on these two stars with another system would be usefule for isolating this issue. Is it the instruments (multiple SSP3 photometers), or is it the star B and V catalog magnitudes?
Jim
Oops. I meant to type a B-V difference of 2.0 to represent some of the worst of the red variables.
Yes, eps-V for my SSP-3 = -0.036.
Test comparisons already exist in the data base. Look for a variable such as P Cyg that is getting currents observations, and see how well your own observation fits for light curve.
The desired precision of +/-0.02 mag applies, I believe, to the internal consistency of the observations. Means of three measurements with a standard deviation of +/-0.02 are accepted, bigger values are rejected. Of course, the program also wants data that is within a hundredth or two of the actual differential magnitude of the comp and variable. But the transformation coefficient plays only a small part. In the extreme cases (comp-var B-V difference of more than 1.0 magnitude) a 10 percent error in the eps-V would change the final result by less than 0.01.
Thanks! I made a typo myself in an earlier post, where I said the Aquarius Red star was 6.015 (should have been 6.01).
I maybe suffering from confusion regarding precision and accuracy, but I interpret an accuracy of 0.02 magnitude as meaning the differential magnitude measurement is no more than 0.02 from the "true" value. If my eV compensation introduces a shift of 0.01, then I have used up half of the error budget before considering the measurement uncertainty. Perhaps the web page really means precision instead of accuracy in this context:
"Integrity of our database requires that observations be accurate to within about 0.02 magnitudes. "
http://www.aavso.org/photoelectric-photometry-pep-observing-program-aav…
Hi Tom,
There's a lot to respond to here. To discuss the issues you raised, we think in general you're correct in what you talked about. Realize the methodology used for the PEP program was developed several decades ago, and has produced very consistent results since that time since all of the observers participating in the PEP-V program were using essentially the same observing method. It is not perfect, but it produces truly good and consistent photometry.
Some specifics:
1) The error budget of 0.02 is an overall goal, but given uncertainties in calibration the *total* error budget of random + systematic is likely higher, but probably not as high as 0.05; we agree with David that for most stars in the PEP program, systematics are probably much lower (on order of 5 mmag). We do stress that the random errors should be around 0.02. Note that we also historically use an assumed mean extinction value regardless of location or conditions, and of course we assume a mean color for the variable star, even though many change color through their light curve. These are not bad assumptions, but they will also increase the systematic uncertainties.
2) Catalog photometry: the errors you're seeing in your epsilon(V) calculation using different catalog values are for the most part unavoidable because there really is no "true" catalog value for these stars. The data that were given on the page of red-blue pairs were *mean* values adapted from the General Catalogue of Photometric Data, and are derived from a number of photometric sources. To our knowledge, none of the stars among the red-blue pairs are fundamental standards (e.g. Cousins or Landolt stars). For example, if you look at the data for the Aquarius pair from GCPD, the mean photometry shows uncertainties in the V-band magnitude of 5 and 8 millimags for HR 8451 and 8453 respectively. So from the V-band catalog uncertainties *alone* you have an uncertainty in epsilon(V) of nearly (10 millimag) / (delta(B-V)). And that's before you make a single measurement.
3) Different results for different pairs: again, this probably goes to the issue of there not being a "true" catalog value for these stars. In addition, every star has a unique spectral energy distribution (SED), and so wide-band filters average the SED over their bandpass. Each pair will then give a unique measurement that may be slightly different from the standard system from pair to pair. We think you suggested this, but it would probably be more rigorous to determine epsilon(V) based on measurements of several pairs of stars, under the assumption that the uncertainties in catalog magnitudes and SED measurement will decline by the square root of the number of pairs that you use. Again, the current procedure was developed early on in the PEP program to yield a scientifically useful balance between statistical rigor and ease of measurement.
What we would suggest is the following:
* Use the photometric data from the web page, rather than what is printed on the PEP finder charts. That is what we intended the observers would do. Those values may not be "true" to better than 0.01 mag, but at least all observers will be on that same system.
* If you want to get a more rigorous measure of your epsilon(V), measure as many red-blue pairs as is feasible given your location and the time of year. If you're able to measure three or more, look at the standard deviation of the resulting epsilon(V) values. Our suspicion is that you'll find they're all within 5-10 percent, which will again translate to an improved calibration at the level of a few percent of the color difference of the variable and comparison star (which, as David said, is small for most stars).
Matthew & Arne
Matt:
Thanks for the detailed response! I don't mean to cause trouble, but I have been very concerned about my eV scatter. I have, in fact, made multiple eV observations. Some have come out absurd, which may be inexperience on my part, but I offer, below, the results I trust most. All reductions are based upon web page magnitudes.
Aqr -0.040
And -0.045
Ori -0.061
Ori -0.066
Per -0.037
LMi -0.068
LMi -0.068
The distribution appears to be bi-modal, with peaks at roughly -0.041 and -0.066.
Tom
Hi Tom,
Did you follow the procedure on the web page, measuring the red/blue pair(s) about 8 times each? If so, what was the standard deviation of the measures for each star? These were done on pristine nights in Bend? Interesting distribution. I'm a little concerned about the ones that you say came out absurd, as that gives me less confidence that these others, which are "cherry picked", are also correct.
With BSM (CCD, not PEP), I measure ~100 stars over the sky, determining extinction at the same time as the transformation coefficients. Based on several nights, my results have +/- 0.006 in the slope. Yours have +/- 0.01 or so, which doesn't appear to be that bad to me. You may be worried about a seemingly systematic difference when in reality it is noise.
Arne
Arne:
My absurd result was for Pegasus: -0.1330. I have a value for Ophiuchus of -0.0746 which was computed from only 5 measurements of the blue star due to a mistake on my part. Those are the two values I do not trust. Aside from Oph, all reductions were based on the standard eV series of deflections.
Below are the mean v(0) values and standard deviations of v(0)
Oph -0.4181 0.0042
Aqr 0.2209 0.0031
Peg 0.1416 0.0089
And -0.0419 0.0010
Per 1.4180 0.0050
Ori 1.2140 0.0067
Ori 1.2073 0.0051
LMi 0.3076 0.0036
LMi 0.3085 0.0110
On the Perseus and one Leo Minor observation I had to go beyond one hour from the meridian, but both these pairs were at high sky altitude, and my telescope is at 6300 feet elevation. A few observations were made at less than pristine transparency (as forecast by clear sky clock); I assumed they would be usable because of the elevation.
I have never made an extinction observation, can you explain the meaning of "slope" in that context?
Tom
Hi Tom,
By "slope" I mean the value of the coefficient, as this is usually determined, when using several stars, by the least-squares fit of (V-v) vs (B-V). For a pair of stars, this is uniquely determined by Doug Hall's technique.
Your conditions sound good enough, and you were following the procedure correctly, so I'd say that you are dealing with small-number statistics (where it is not two peaks, but a continuum, with missing values), there was something occurring during your observations (such as some taken on one night, and some taken on another night with different extinction), or the catalog magnitude/color values have some systematic error included. In any case, the results are "good enough" if you just average your calculated values.
Arne
SAO 91253 is ups Peg at 23 25 22.78350 +23 24 14.7606
Kharchenko V=4.422 +/- 0.002 (B-V) = 0.616 +/- 0.004
Mermilliod gives it 4.407 +/-0.013 (B-V) = 0.610 +/- 0.006
Listed as 4.40 in the AAVSO table
SAO 91186 is tau Peg at 23 20 38.24188 +23 44 25.2098
it is a delta Scuti star, with full amplitude 0.02mag
Kharchenko V=4.581 +/- 0.002 (B-V) = 0.195 +/- 0.002
Mermilliod V = 4.592 +/- 0.007 (B-V) = 0.171 +/- 0.010
Listed as 4.60 in the AAVSO table
delta (B-V) from the literature is thus 0.430, compared with the AAVSO table value of 0.44
I don't think that this star pair has changed significantly over the years. I could remeasure with BSM, but it would take a dozen nights or so to reach this precision. If I were to guess, I'd say that there may be some companions that are being included on the diode due to its large effective "aperture" size. The color difference is pretty small for transformation coefficient determination (the smallest difference in the entire table). I'd weight any result from it less than from other pairs.
Arne
Arne,
Thanks for the reply. As you point out the sensitivity increases as the delta(B-V) goes down. The catalog data you present sheds some light on the issue. The delta V in the AAVSO table is 0.2. The values you give range from 0.159 +/- 0.004 for Kharchenko and 0.185 +/- 0.020 for Mermillod.
In my case a value of delta V of 0.162 gives me the same value for epsilon V as I get using my measurements of the Hercules pair. Very close to the values of Kharchenko and Mermilod when weighted statistically by their errors.
With such a small delta(B-V) the v and V values must be pretty accurate to give a reasonable answer. If I want my epsilon V to be accurate to +/-20%, I would need both the deltaV and deltav to be accurate to about 0.005 mag. Other pairs with larger delta(B-V) will be proportionately less sensitive.
I have some data from last night on the Andromeda pair which I still need to reduce. With a delta(B-V) of -0.77 it is still not one of the best pairs. Aquarius is a bit better at -0.985, but the data I took last night on that pair is unlikely to have the required accuracy as it was very close to the full moon, resulting in sky readings almost 10 times my normal background.
In any case the catalog data you provided clears up the issue at least for my measurement system.